AnimationBench - Behavior Consistency: leaderboard
Metric: IP behavior consistency: whether the character's signature traits, environmental interaction and action logic match its lore (%), animation-style image-to-video generation from the AnimationBench close-set suite (fixed source images and prompts, about 30 videos per dimension per model), scored by Qwen3-VL-Max yes/no question sets aggregated per video; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Veo 3.1 | 84.72 |
| 2 | Kling 2.6 | 83.33 |
| 3 | Wan2.2-I2V-A14B | 80.09 |
| 4 | Seedance 2.0 | 78.44 |
| 5 | Seedance 1.0 Pro | 78.24 |
| 6 | Sora 2 Pro | 77.31 |
| 7 | FramePack | 61.3 |
| 8 | HunyuanVideo-1.5 | 47.91 |
Interactive version: theaggregate.ai/benchmark?slug=animationbench-behavior-consistency · How It Works · Data refreshed daily, snapshot 2026-10-07.